Re: [RFA] 12843

[Tom Tromey <tromey@redhat.com>]

Tom> (I'd like -break-insert to avoid linespecs completely, which would be a
Tom> big improvement IMO, ...

André> From an MI client's point of view, passing all location information as 
André> a single argument is neither wanted nor needed.

Yeah.  I am not sure why it works this way, as it seems to be against
the whole idea of MI.

André> or even require everything to be quoted as in
André>  25-break-insert  --file "some thing.cpp" --line "794"

There's no need; MI already defines an input quoting syntax that is
sufficient.

Tom> ... but of course we still have to worry about compatibility.)

André> Just using new flags for the parameters should do the trick in this case.

André> In general, in the past, differences between different versions
André> of gdb have been large enough to require modifications in
André> consumer code anyway (both to take advantage of new features, and
André> to handle incompatibilities in input syntax), so requiring 100%
André> "feature" compatibility _just for the sake of it_ is unlikely to
André> be helpful. At least I would prefer a syntax with simple, uniform
André> quoting rules over ad-hoc solutions tied to a specific OS or file
André> system.

For MI, I don't think there is any issue.  We can just add options to
-break-insert.  (I don't know of anybody specifically working on this,
but I will at least put it in bugzilla.)

My view is that it is best to be compatible with existing "reasonable"
practice when possible, where "reasonable" means something akin to "has
actually happened".

In this case I think we can simplify linespec lexing without unduly
breaking things.

Tom> I think adopting these rules will make some of my ambiguous linespec
Tom> changes simpler. 

André> I wonder whether it would be possible to just leave the current syntax
André> unchanged, and introduce a new, better behaved syntax and use some
André> global setting for toggling between them. So everybody concerned about
André> compatibility would not notice a change, and the others would put 
André>   'set breakpoint syntax 2011'  [or whatever]
André> in their .gdbinit and could use the new way.

My first reaction is against this, but I don't have a particularly good
explanation for that.  I will think about it.

Tom